{
  "id": 12523,
  "title": "Unsupervised pre-training on testing set",
  "url": "/competitions/datasciencebowl/discussion/12523",
  "author_name": "",
  "post_date": "2015-02-15T22:22:29.050Z",
  "votes": null,
  "comment_count": 1,
  "views": 884,
  "content": "<p>Has anyone had any luck with performing unsupervised pre-training on the testing set? I have tried using an denoising autoencoder and so far my performance has not improved at all, basically it did nothing. Just wondering if anyone has had any luck before I devote a bunch more time into this.&nbsp;</p>",
  "messages": [
    {
      "id": "64292",
      "postDate": "02/15/2015 22:22:29",
      "content": "<p>Has anyone had any luck with performing unsupervised pre-training on the testing set? I have tried using an denoising autoencoder and so far my performance has not improved at all, basically it did nothing. Just wondering if anyone has had any luck before I devote a bunch more time into this.&nbsp;</p>",
      "rawMarkdown": "",
      "votes": null
    },
    {
      "id": "64303",
      "postDate": "02/16/2015 02:23:06",
      "content": "<p>My feeling is it will not be worth doing - I think the fraction of the test set which is real is comparable to or even smaller than the training set, so there&nbsp;won't be much gain over data augmentation. Would be great&nbsp;to hear otherwise though.</p>",
      "rawMarkdown": "",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 64303,
      "author_name": "emolson",
      "author_url": "",
      "post_date": "02/16/2015 02:23:06",
      "content": "<p>My feeling is it will not be worth doing - I think the fraction of the test set which is real is comparable to or even smaller than the training set, so there&nbsp;won't be much gain over data augmentation. Would be great&nbsp;to hear otherwise though.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "64292": "",
    "64303": ""
  },
  "source": "meta"
}